Predicting the functional state of protein kinases using interpretable graph neural networks from sequence and structural data.

Protein kinases are central to cellular activities and are actively pursued as drug targets for several conditions including cancer and autoimmune diseases. Despite the availability of a large structural database for kinases, methodologies to elucidate the structure-function relationship of these proteins (without manual intervention) are lacking. Such techniques are essential in structural biology and to accelerate drug discovery efforts. Here, we implement an interpretable graph neural network (GNN) framework for classifying the functionally active and inactive states of a large set of protein kinases by only using their tertiary structure and amino acid sequence. We show that the GNN models can classify kinase structures with high accuracy (>97%). We implement the Gradient-weighted Class Activation Mapping for graphs (Graph Grad-CAM) to automatically identify structurally important residues and residue-residue contacts of the kinases without any a priori input. We show that the motifs identified through the Graph Grad-CAM methodology are functionally critical, consistent with the existing kinase literature. Notably, the highly conserved DFG and HRD motifs of the well-known hydrophobic spine are identified by the interpretable framework in addition to some of the lesser known motifs. Further, using Grad-CAM maps as the vector embedding of the protein structures, we identify the subtle differences in the crystal structures among different sub-classes of kinases in the Protein Data Bank (PDB). Frameworks such as the one implemented here, for high-throughput identification of protein structure-function relationships are essential in designing targeted small molecules therapies as well as in engineering new proteins for novel applications.

Computational Design of Low Melting Eutectics of Molten Salts: A Combined Machine Learning and Thermodynamic Modeling Approach.

We develop a computational framework combining thermodynamic and machine learning models to predict the melting temperatures of molten salt eutectic mixtures (Teut). The model shows an accuracy of ∼6% (mean absolute percentage error) over the entire data set. Using this approach, we screen millions of combinatorial eutectics ranging from binary to hexanary, predict new mixtures, and propose design rules that lead to low Teut. We show that heterogeneity in molecular sizes, quantified by the molecular volume of the components, and mixture configurational entropy, quantified by the number of mixture components, are important factors that can be exploited to design low Teut mixtures. While predicting eutectic composition with existing techniques had proved challenging, we provide some preliminary models for estimating the compositions. The high-throughput screening technique presented here is essential to design novel mixtures for target applications and efficiently navigate the vast design space of the eutectic mixtures.

Microstructural mechanisms of hysteresis and transformation width in NiTi alloy from molecular dynamics simulations.

Martensitic transformations in shape memory alloys are often accompanied by thermal hysteresis, and engineering this property is of prime scientific interest. The martensitic transformation can be characterized as thermoelastic, where the extent of the transformation is determined by a balance between thermodynamic driving force and stored elastic energy. Here we used molecular dynamics simulations of the NiTi alloy to explore hysteresis-inducing mechanisms and thermoelastic behavior by progressively increasing microstructural constraints from single crystals to bi-crystals to polycrystals. In defect-free single crystals, the austenite-martensite interface moves unimpeded with a high velocity. In bi-crystals, grain boundaries act as significant obstacles to the transformation and produce hysteresis by requiring additional nucleation events. In polycrystals, the transformation is further limited by the thermoelastic balance. The stored elastic energy can be converted to mechanisms of non-elastic strain accommodation, which also produce hysteresis. We further demonstrated that the thermoelastic behavior can be controlled by adjusting microstructural constraints.

A 3D printable alloy designed for extreme environments.

Multiprincipal-element alloys are an enabling class of materials owing to their impressive mechanical and oxidation-resistant properties, especially in extreme environments1,2. Here we develop a new oxide-dispersion-strengthened NiCoCr-based alloy using a model-driven alloy design approach and laser-based additive manufacturing. This oxide-dispersion-strengthened alloy, called GRX-810, uses laser powder bed fusion to disperse nanoscale Y2O3 particles throughout the microstructure without the use of resource-intensive processing steps such as mechanical or in situ alloying3,4. We show the successful incorporation and dispersion of nanoscale oxides throughout the GRX-810 build volume via high-resolution characterization of its microstructure. The mechanical results of GRX-810 show a twofold improvement in strength, over 1,000-fold better creep performance and twofold improvement in oxidation resistance compared with the traditional polycrystalline wrought Ni-based alloys used extensively in additive manufacturing at 1,093 °C5,6. The success of this alloy highlights how model-driven alloy designs can provide superior compositions using far fewer resources compared with the 'trial-and-error' methods of the past. These results showcase how future alloy development that leverages dispersion strengthening combined with additive manufacturing processing can accelerate the discovery of revolutionary materials.

Modeling Singlet Oxygen-Induced Degradation Pathways Including Environmental Effects of 1,2-Dimethoxyethane in Li-O2 Batteries through Density Functional Theory.

We employ density functional theory (DFT) to examine reaction mechanisms involving singlet oxygen 1Δg (1O2) and 1,2-dimethoxyethane (DME) to probe potential parasitic reactions occurring in Li-O2 batteries. First, we investigate the attack of 1O2 on the ethylene group (-CH2-CH2-) to form H2O2 and a C-C double bond in a single step. Second, we look at hydroperoxide formation that occurs via a two-step mechanism. We employ an implicit solvent model, Li+ coordination, and external electric fields to model the complex electrolyte environment near the cathode of a Li-O2 battery. The initial barriers for these reactions are decreasing functions of the dielectric constant of the implicit solvent model as well as the strength of the electric field. These initial barriers range between 17 and 26 kcal mol-1 for large dielectric constants and in the presence of electric fields. We discuss the implications of these results on ether-based electrolytes for Li-O2 batteries.

Nature of the Electrical Double Layer on Suspended Graphene Electrodes.

The structure of interfacial water near suspended graphene electrodes in contact with aqueous solutions of Na2SO4, NH4Cl, and (NH4)2SO4 has been studied using confocal Raman spectroscopy, sum frequency vibrational spectroscopy, and Kelvin probe force microscopy. SO42- anions were found to preferentially accumulate near the interface at an open circuit potential (OCP), creating an electrical field that orients water molecules below the interface, as revealed by the increased intensity of the O-H stretching peak of H-bonded water. No such increase is observed with NH4Cl at the OCP. The intensity of the dangling O-H bond stretching peak however remains largely unchanged. The degree of orientation of the water molecules as well as the electrical double layer strength increased further when positive voltages are applied. Negative voltages on the other hand produced only small changes in the intensity of the H-bonded water peaks but affected the intensity and frequency of dangling O-H bond peaks. The TOC figure is an oversimplified representation of the system in this work.

Predicting the Ion Desolvation Pathway of Lithium Electrolytes and Their Dependence on Chemistry and Temperature.

To better understand the influence of electrolyte chemistry on the ion-desolvation portion of charge-transfer beyond the commonly applied techniques, we apply free-energy sampling to simulations involving diethyl ether (DEE) and 1,3-dioxoloane/1,2-dimethoxyethane (DOL/DME) electrolytes, which display bulk solvation structures dominated by ion-pairing and solvent coordination, respectively. This analysis was conducted at a pristine electrode with and without applied bias at 298 and 213 K to provide insights into the low-temperature charge-transfer behavior, where it has been proposed that desolvation dominates performance. We find that, to reach the inner Helmholtz layer, ion-paired structures are advantageous and that the Li+ ion must reach a total coordination number of 3, which requires the shedding of 1 species in the DEE electrolyte or 2-3 species in DOL/DME. This work represents an effort to predict the distinct thermodynamic states as well as the most probable kinetic pathways of ion desolvation relevant for the charge transfer at electrochemical interphases.

Coarse-Grained Dynamically Accurate Simulations of Ionic Liquids: [pyr14][TFSI] and [EMIM][BF4].

In this work, coarse-grained (CG) models for two different sets of ionic liquids were developed from atomistic molecular dynamics (MD) reference systems, expanding their system size and time duration capabilities. The bonded force field of the CG systems was built using harmonic oscillator potential (HOP) fitting, while the nonbonded force field was generated with the multiscale coarse-graining (MS-CG) approach based on force matching. The dynamics of each system were corrected using the probability distribution function-based coarse-grained molecular dynamics (PDF-based CGMD) method. The structure and dynamics of each system were proven to match reference system data at two temperature scales. CG models and force fields for these liquids were developed to exemplify a general purpose methodology for producing MD results of ionic liquids and other fluids with accurate structural as well as dynamic properties. As an application, developed ionic liquids CG models were then applied to study vacuum-interface interaction. Density profile results of vacuum-interface exposure show significant deviation from bulk behavior. At the interface, multilayer ordering of ionic liquids is predicted to be similar to those observed from an experimental work. This ordering is intensified by decreasing temperature and use of the PDF-based CGMD method as opposed to conventional CG methods.

Discovery of novel Li SSE and anode coatings using interpretable machine learning and high-throughput multi-property screening.

All-solid-state batteries with Li metal anode can address the safety issues surrounding traditional Li-ion batteries as well as the demand for higher energy densities. However, the development of solid electrolytes and protective anode coatings possessing high ionic conductivity and good stability with Li metal has proven to be a challenge. Here, we present our informatics approach to explore the Li compound space for promising electrolytes and anode coatings using high-throughput multi-property screening and interpretable machine learning. To do this, we generate a database of battery-related materials properties by computing [Formula: see text] migration barriers and stability windows for over 15,000 Li-containing compounds from Materials Project. We screen through the database for candidates with good thermodynamic and electrochemical stabilities, and low [Formula: see text] migration barriers, identifying promising new candidates such as [Formula: see text]N, [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text], among others. We train machine learning models, using ensemble methods, to predict migration barriers and oxidation and reduction potentials of these compounds by engineering input features that ensure accuracy and interpretability. Using only a small number of features, our gradient boosting regression models achieve [Formula: see text] values of 0.95 and 0.92 on the oxidation and reduction potential prediction tasks, respectively, and 0.86 on the migration barrier prediction task. Finally, we use Shapley additive explanations and permutation feature importance analyses to interpret our machine learning predictions and identify materials properties with the largest impact on predictions in our models. We show that our approach has the potential to enable rapid discovery and design of novel solid electrolytes and anode coatings.

Reaction of Singlet Oxygen with the Ethylene Group: Implications for Electrolyte Stability in Li-Ion and Li-O2 Batteries.

Recent experimental and computational evidence indicates that singlet oxygen (1O2) attacks the ethylene group (-CH2-CH2-) in ethylene carbonate (EC) leading to degradation in Li-ion batteries employing EC as the electrolyte solvent [J. Phys. Chem. A 2018, 122, 8828-8839]. Here, we employ computational quantum chemistry to explore this mechanism in detail for a large set of organic molecules. Benchmark calculations comparing density functional theory to the complete active space second-order perturbation theory and internally contracted multireference configuration interaction indicate that the M11 functional adequately captures trends in the transition-state energies for this mechanism. Based on our results, we recommend that solvents which include the ethylene group should be avoided in Li-ion and Li-O2 batteries where 1O2 is generated unless neighboring functional groups raise the reaction barrier to avoid this decomposition pathway.

Anion-Assisted Delivery of Multivalent Cations to Inert Electrodes.

To understand and control key electrochemical processes-metal plating, corrosion, intercalation, etc.-requires molecular-scale details of the active species at electrochemical interfaces and their mechanisms for desolvation from the electrolyte. Using free energy sampling techniques we reveal the interfacial speciation of divalent cations in ether-based electrolytes and mechanisms for their delivery to an inert graphene electrode interface. Surprisingly, we find that anion solvophobicity drives a high population of anion-containing species to the interface that facilitate the delivery of divalent cations, even to negatively charged electrodes. Our simulations indicate that cation desolvation is greatly facilitated by cation-anion coupling. We propose anion solvophobicity as a molecular-level descriptor for rational design of electrolytes with increased efficiency for electrochemical processes limited by multivalent cation desolvation.

Proton Abstraction from DME n···X+ by OH-, O2-, and XO2-, for X = Li, Na, and K: Implications for Li-O2 Batteries.

The abstraction of a proton by OH-, O2-, and XO2- from DME n···X+, where X is Li, Na, or K, is studied using density functional theory. Both the gas phase and the solution phase are studied. In general, when explicit solvent molecules are added, the difference between the gas-phase and solution results becomes rather small. While the DME n···X+ binding energies differ significantly for various alkali cations, the reaction energies and transition-state energies are far less sensitive to the choice of an alkali cation. XO2- has a lower barrier height than OH-, which, in turn, has a lower barrier height than O2-. The reaction energies follow the same trends.

Polarizable Molecular Dynamics and Experiments of 1,2-Dimethoxyethane Electrolytes with Lithium and Sodium Salts: Structure and Transport Properties.

The structure and transport properties of electrolyte solutions of 1,2-dimethoxyethane (DME) having salts of Li+ with bis(trifluoromethanesulfonyl)imide ([TFSI]-) or Na+ with [TFSI]- are investigated with polarizable molecular dynamics and experiments. Polarizable force fields for Li+ and Na+ with DME and [TFSI]- were developed based on quantum chemistry calculations, ab initio molecular dynamics simulations, and thermodynamic liquid-state properties. Simulation results for density, viscosity, self-diffusion coefficient, and conductivity of the electrolytes all agree well with the trends and magnitudes of available experimental data for a wide range of salt concentrations. As the concentration of salt increases, the electrolytes become more viscous and molecular species become less mobile. Ionic conductivity does not change monotonically with salt concentration and exhibits a maximum between 0.5 and 1.0 M for both Li[TFSI] and Na[TFSI] electrolytes. Comparatively, both cations are solvated by 5-6 DME or [TFSI]- oxygen atoms and exhibit similar diffusivities and conductivities. The solvation shell of Na+ is found to be more weakly bound and to have a lower binding residence time than that of Li+. The transport of Li+ therefore is more vehicular, through the motion of the solvation shell, while the transport of Na+ is based more on exchange, through the replacement of solvating species. The atomistic insight provided by this work can be used as the basis for future rational design of improved electrolyte solvents for lithium-oxygen, sodium-oxygen, and lithium-sulfur batteries.

Phenolic Polymer Solvation in Water and Ethylene Glycol, I: Molecular Dynamics Simulations.

Interactions between pre-cured phenolic polymer chains and a solvent have a significant impact on the structure and properties of the final postcured phenolic resin. Developing an understanding of the nature of these interactions is important and will aid in the selection of the proper solvent that will lead to the desired final product. Here, we investigate the role of the phenolic chain structure and the solvent type on the overall solvation performance of the system through molecular dynamics simulations. Two types of solvents are considered: ethylene glycol (EGL) and H2O. In addition, three phenolic chain structures are considered, including two novolac-type chains with either an ortho-ortho (OON) or an ortho-para (OPN) backbone network and a resole-type (RES) chain with an ortho-ortho network. Each system is characterized through a structural analysis of the solvation shell and the hydrogen-bonding environment as well as through a quantification of the solvation free energy along with partitioned interaction energies between specific molecular species. The combination of simulations and the analyses indicate that EGL provides a higher solvation free energy than H2O due to more energetically favorable hydrophilic interactions as well as favorable hydrophobic interactions between CH element groups. In addition, the phenolic chain structure significantly affects the solvation performance, with OON having limited intermolecular hydrogen-bond formations, while OPN and RES interact more favorably with the solvent molecules. The results suggest that a resole-type phenolic chain with an ortho-para network should have the best solvation performance in EGL, H2O, and other similar solvents.

Phenolic Polymer Solvation in Water and Ethylene Glycol, II: Ab Initio Computations.

Ab initio techniques are used to study the interaction of ethylene glycol and water with a phenolic polymer. The water bonds more strongly with the phenolic OH than with the ring. The phenolic OH groups can form hydrogen bonds between themselves. For more than one water molecule, there is a competition between water-water and water-phenolic interactions. Ethylene glycol shows the same effects as those of water, but the potential energy surface is further complicated by CH2-phenolic interactions, different conformers of ethylene glycol, and two OH groups on each molecule. Thus, the ethylene glycol-phenolic potential is more complicated than the water-phenolic potential. The results of the ab initio calculations are compared to those obtained using a force field. These calibration studies show that the water system is easier to describe than the ethylene glycol system. The calibration studies confirm the reliability of force fields used in our companion molecular dynamics study of a phenolic polymer in water and ethylene solutions.

Finite Temperature Properties of NiTi from First Principles Simulations: Structure, Mechanics, and Thermodynamics.

We present a procedure to determine temperature-dependent thermodynamic properties of crystalline materials from density functional theory molecular dynamics (DFT-MD). Finite temperature properties (structural, thermal, and mechanical properties) of the phases (ground state monoclinic B33, martensitic B19', and austenitic B2) of the shape memory alloy NiTi are investigated. Fluctuation formulas and numerical derivatives are used to evaluate mechanical and thermal properties. A modified version of thermodynamic upsampling is used to enable simulations with lower DFT convergence thresholds. In addition, a thermodynamic integration expression is developed to compute free energies from isobaric DFT-MD simulations that accounts for volume changes. Structural parameters, elastic constants, volume expansion, and specific heats as a function of temperature are evaluated. Phase transitions between B2 and B19' and between B19' and B33 are characterized according to their thermal energy, entropy, and free energy differences as well as their latent heats. Anharmonic effects are shown to play a large role in both stabilizing the austenite B2 phase as well as suppressing the martensitic phase transition. The quasiharmonic approximation to the free energy results in large errors in estimating the martensitic transition temperature by neglecting these large anharmonic components.

Evaluation of molecular dynamics simulation methods for ionic liquid electric double layers.

We investigate how systematically increasing the accuracy of various molecular dynamics modeling techniques influences the structure and capacitance of ionic liquid electric double layers (EDLs). The techniques probed concern long-range electrostatic interactions, electrode charging (constant charge versus constant potential conditions), and electrolyte polarizability. Our simulations are performed on a quasi-two-dimensional, or slab-like, model capacitor, which is composed of a polarizable ionic liquid electrolyte, [EMIM][BF4], interfaced between two graphite electrodes. To ensure an accurate representation of EDL differential capacitance, we derive new fluctuation formulas that resolve the differential capacitance as a function of electrode charge or electrode potential. The magnitude of differential capacitance shows sensitivity to different long-range electrostatic summation techniques, while the shape of differential capacitance is affected by charging technique and the polarizability of the electrolyte. For long-range summation techniques, errors in magnitude can be mitigated by employing two-dimensional or corrected three dimensional electrostatic summations, which led to electric fields that conform to those of a classical electrostatic parallel plate capacitor. With respect to charging, the changes in shape are a result of ions in the Stern layer (i.e., ions at the electrode surface) having a higher electrostatic affinity to constant potential electrodes than to constant charge electrodes. For electrolyte polarizability, shape changes originate from induced dipoles that soften the interaction of Stern layer ions with the electrode. The softening is traced to ion correlations vertical to the electrode surface that induce dipoles that oppose double layer formation. In general, our analysis indicates an accuracy dependent differential capacitance profile that transitions from the characteristic camel shape with coarser representations to a more diffuse profile with finer representations.

Computational and Experimental Study of Li-Doped Ionic Liquids at Electrified Interfaces.

We evaluate the influence of Li-salt doping on the dynamics, capacitance, and structure of three ionic liquid electrolytes, [pyr14][TFSI], [pyr13][FSI], and [EMIM][BF4], using molecular dynamics and polarizable force fields. In this respect, our focus is on the properties of the electric double layer (EDL) formed by the electrolytes at the electrode surface as a function of surface potential (Ψ). The rates of EDL formation are found to be on the order of hundreds of picoseconds and only slightly influenced by the addition of Li-salt. The EDLs of three electrolytes are shown to have different energy storage capacities, which we relate to the EDL formation free energy. The differential capacitance obtained from our computations exhibits asymmetry about the potential of zero charge and is consistent with the camel-like profiles noted from mean field theories and experiments on metallic electrodes. The introduction of Li-salt reduces the noted asymmetry in the differential capacitance profile. Complementary experimental capacitance measurements have been made on our three electrolytes in their neat forms and with Li-salt. The measurements, performed on glassy carbon electrodes, produce U-like profiles, and Li-salt doping is shown to strongly affect capacitance at high magnitudes of Ψ. Differences in the theoretical and experimental shapes and magnitudes of capacitance are rationalized in terms of the electrode surface and pseudocapacitive effects. In both neat and Li-doped liquids, the details of the computational capacitance profile are well described by Ψ-induced changes in the density and molecular orientation of ions in the molecular layer closest to the electrode. Our results suggest that the addition of Li+ induces disorder in the EDL, which originates from the strong binding of anions to Li+. An in-depth analysis of the distribution of Li+ in the EDL reveals that it does not readily enter the molecular layer at the electrode surface, preferring instead to be localized farther away from the surface in the second molecular layer. This behavior is validated through an analysis of the free energy of Li+ solvation as a function of distance from the electrode. Free energy wells are found to coincide with localized concentrations of Li+, the depths of which increase with Ψ and suggest a source of impedance for Li+ to reach the electrode. Finally, we make predictions of the specific energy at ideal graphite utilizing the computed capacitance and previously derived electrochemical windows of the liquids.

Ab Initio Simulations and Electronic Structure of Lithium-Doped Ionic Liquids: Structure, Transport, and Electrochemical Stability.

Density functional theory (DFT), density functional theory molecular dynamics (DFT-MD), and classical molecular dynamics using polarizable force fields (PFF-MD) are employed to evaluate the influence of Li(+) on the structure, transport, and electrochemical stability of three potential ionic liquid electrolytes: N-methyl-N-butylpyrrolidinium bis(trifluoromethanesulfonyl)imide ([pyr14][TFSI]), N-methyl-N-propylpyrrolidinium bis(fluorosulfonyl)imide ([pyr13][FSI]), and 1-ethyl-3-methylimidazolium boron tetrafluoride ([EMIM][BF4]). We characterize the Li(+) solvation shell through DFT computations of [Li(Anion)n]((n-1)-) clusters, DFT-MD simulations of isolated Li(+) in small ionic liquid systems, and PFF-MD simulations with high Li-doping levels in large ionic liquid systems. At low levels of Li-salt doping, highly stable solvation shells having two to three anions are seen in both [pyr14][TFSI] and [pyr13][FSI], whereas solvation shells with four anions dominate in [EMIM][BF4]. At higher levels of doping, we find the formation of complex Li-network structures that increase the frequency of four anion-coordinated solvation shells. A comparison of computational and experimental Raman spectra for a wide range of [Li(Anion)n]((n-1)-) clusters shows that our proposed structures are consistent with experiment. We then compute the ion diffusion coefficients and find measures from small-cell DFT-MD simulations to be the correct order of magnitude, but influenced by small system size and short simulation length. Correcting for these errors with complementary PFF-MD simulations, we find DFT-MD measures to be in close agreement with experiment. Finally, we compute electrochemical windows from DFT computations on isolated ions, interacting cation/anion pairs, and liquid-phase systems with Li-doping. For the molecular-level computations, we generally find the difference between ionization energy and electron affinity from isolated ions and interacting cation/anion pairs to provide upper and lower bounds, respectively, to experiment. In the liquid phase, we find the difference between the lowest unoccupied and highest occupied electronic levels in pure and hybrid functionals to provide lower and upper bounds, respectively, to experiment. Li-doping in the liquid-phase systems results in electrochemical windows little changed from the neat systems.

Structure and energetics of Li(+)-(BF4(-))n, Li(+)-(FSI(-))n, and Li(+)-(TFSI(-))n: ab initio and polarizable force field approaches.

The Li(+)-BF4(-) and BF4(-)-BF4(-) interactions are studied using second order perturbation theory (MP2) and coupled cluster singles and doubles approach, including the effect of connected triples, CCSD(T). The MP2 and CCSD(T) results are in excellent agreement. Using only the MP2 approach, the interactions of Li(+) with bis(trifluoromethane)sulfonimide anion (TFSI) and Li(+) with bis(fluorosulfonyl)imide anion (FSI) are studied. The results of these high level calculations are compared with density functional theory (DFT) calculations for a variety of functionals and with the APPLE&P force field. The B3LYP approach well reproduces the accurate calculations using both a small and large basis set. The M06 and M06L functionals in the larger basis set are in good agreement with the high level calculations. While the APPLE&P force field does not outperform the best functionals, the APPLE&P results agree better with the accurate results than do some of the functionals tested.